Independent AI Research · India

0labs

Coming soon.

We're building a new kind of language model — one that adapts how deeply it thinks based on how hard the problem is. Simple questions get fast answers. Complex reasoning gets more compute. Automatically.

Coming:Sky v2.0 (11B)Research Paper

What's Coming

A language model with adaptive depth, and the research behind it.

labs

Sky v2.0 — 11B Parameters

A language model that doesn't waste compute. Every layer decides how many steps each token needs — finishing fast on easy tokens and thinking deeper on hard ones. Expanded from 4B to 11B through a proprietary architecture.

Architecture Paper

Coming 2026

Full architecture details, mathematical formalization, and adaptive depth analysis.

Evaluation & Benchmarks

Coming 2026

Code generation results, reasoning benchmarks, depth behavior analysis.

The Team

Built by one.

0labs is run entirely by Atharvsinh Jadav — one researcher building adaptive AI from Gujarat, India, on a single GPU.

No institutional funding. No compute cluster. Just a clear idea about how language models should allocate their computation, and the engineering to make it real.

100

ADAPTIVE DEPTH · SKY V2.0 · 11 BILLION PARAMETERS · MADE IN INDIA•